MODEL BASED DESIGN AND SDR

Mansour Ahmadian , Zhila (Jila) Nazari ,Nory Nakhaee ,Zoran Kostic

Sundance Multiprocessing Technology Ltd, Chesham, UK (mansour.a@sundance.com) School of Engineering and Electronics. University Of Edinburgh, Scotland, UK (z.j.nazari@ed.ac.uk) Sundance Digital Signal Processing Inc, Reno, USA (nory.n@sundance.com) The MathWorks, Natick, USA (wireless.sdr@mathworks.com)

Keywords: SDR, Model Based Design, DSP, Embedded software development

Abstract
This paper focuses on the major phases present in the development of an SDR system: design; simulation; code generation and verication. The paper will illustrate, through an example, the use of Model Based Design methodology and tools to integrate the major development phases into one continuous design cycle. Advanced system design concepts including, simulation, code generation, hardware in the loop testing will be presented. Implementation of an F M 3 T R (Future MultiBand, Multi-Waveform, Modular, Tactical Radio) modulatordemodulator will be discussed and presented.

Fig. 1. Simplied presentation of traditional system design methodology.

1 Introduction
In a traditional development, usually a system engineer denes the overall system specication and presents it as a design document to software engineers, who will have the task of implementing those ideas into a fully working solution. However, the main problem with this approach is the fact that in most cases the ideas presented by the system engineer via the specication document may widely differ to the implemented software. Even the most detailed and diligently prepared type of documentation may not always guarantee that the design document generated by the system engineers would be fully undestood and accurately understood and interpreted correctly by the implementing software programmers. The new emerging technique to solve this problem is model based design [7], In this technique, system engineers design and simulate the model using a model based design tool and he present the ideas behind the solution as a working model of the system . Some of the sophisticated model based design tools, have advanced features for automatic generation of source code that would mimic the simulated system and hence cutting time to market and reducing development costs. In recent years model based design methodology has become the preferred method for designing, modelling and simulating complex dynamic systems [9] [5]. The traditional system design methodology, which is shown in Fig.1, is rather time consuming and error prone in nature. In this simple model, the design and development of a system and its introduction to the market consists of 4 separate steps.

The rst phase involves the denition and design of the objective system. This is the usually carried out by the system designers, who dene the system specication and presents it as a document to software designers. This stage can have the highest risk as it is the biggest source of error since different code developers may interpret these denitions differently. The next step is the implementation part. At this stage the developers will generate suitable software code in a selected language to implement the system. For embedded systems, the preferred programming language is C. If the system denitions are not clear, the programmers would refer to system designer for clarication. Even after these clarications, it is always possible that the developed system may be different from the one that system designers had in mind. After code development, the next step is testing. It is very common that the test result will be different from what the programmers or system designer had in their minds or specied. If the tests show that there are mistakes in the system design, then system designer should revise his design and pass it to code developers to implement the revised system. Of course it can also be the case that the implementation is not a true representation of the specication and hence the test results show errors. In this case the programmer has to make sure he has a better understanding of the specication and either correct the algorithm used or simply debug the application. Either of these processes may be repeated several times. To complicate the matter even further one can also consider the case where the test procedures or implementation are wrong and thus giving erroneous results. The main problem of this approach is the fact that in most

Fig. 2. Model based system design steps. cases the ideas developed by the system engineer/designers and contained in the specication document and the implemented software, may be different. After extensive testing and only when the system designers are convinced that the implemented system is operating correctly and as intended production can start. The complexity of this lengthy process makes project management ineffective and very difcult.

model based design tools as part of the core design process. To demonstrate the viability of this methodology for design, implementation and testing of a complex system, the model based system design technique was used to implement and test an F M 3 T R (Future Multi-Band, Multi-Waveform, Modular, Tactical Radio) modulator-demodulator. The model based design tool that was used for this work was Simulink [14]. Simulink is a well-known tool for modelling, analysing, and simulating a very wide variety of physical and mathematical systems, including those with non-linear elements. As an extension of Matlab, Simulink adds many features specic to dynamic systems while retaining all of Matlabs general purpose functionality and features. Using Simulink, one models a system graphically, sidestepping much of the nuisance associated with conventional programming. SMT6050 from Sundance [4] is used for code generation. SMT6050 uses Real Time Workshop [13] from MathWorks Inc. [3] to generate optimized; embeddable code targeting Sundance hardware. Using SMT6050, generating code for Sundance DSP boards is only a click away. Sundance SDR kit is used as the basic hardware platform for testing the developed code. This kit consists of the following parts: PCI carrier board (310Q): To hold DSP and DAQ board and build a very high speed data communication channel between DSP and PC. DSP board (SMT365G): Texas Instrument TMS320C6416 DSP processor running at 1GHz with 4Mbytes ZBT-RAM. It also contain a Xilinx Virtex II XC2V1000-4 FPGA. DAQ module(SMT370): Two 14-bit ADCs (AD6645) sampling up to 105 MHz. Dual 16-bit TxDAC (AD9777) sampling up to 400 MHz (interpolation). To implement the application rst it was designed and tested with Simulink. When the design phase nished and the system behaviors was proved to meet the specications, the created model was prepared for code generation. To test the generated code, we used Hardware In the Loop (HIL) procedure. In HIL technique, the generated code is run on the real hardware. Simulink inputs the test vectors to the hardware under consideration and collects the system output and processes or displays them to validate the system under test.

2 Model based design

To improve on the above approach and in order to avoid miscomprehension the model-based system design was invented [10] [8]. Model-based design is now an established approach to develop efcient solutions to complex engineering problems [2]. In this method, complicated systems can be created by using mathematical models representing system components and their interactions with their surrounding environment. These models have many applications in the design process, including system simulation, stability analysis, and control algorithm design. By introduction of advanced, automated code-generation technology, another application of these models has become viable. These models can be used as the input to an automatic code generation tool. Advanced, state of the art code generators can produce optimized, embeddable C source codes from these models [1]. Using model based design methodology reduces the no of development stages by combining the design, implementation and test stages in into one process as shown in Fig.2. The reduction of required steps, compared to the traditional method, and complexity will result in better project management and mitigation of product development risks. The systems designed using this approach reach the market faster and end up costing less than the systems designed using the conventional methods.

4 F M 3 T R modulator-demodulator
The implementation of a reference F M 3 T R (Future MultiBand, Multi-Waveform, Modular, Tactical Radio) modulatordemodulator [12] was selected as the SDR system to be developed using model based system methodology. The F M 3 T R was an international cooperative effort between the United States, Germany, France and the United Kingdom to develop a recongurable communication system for ground and airborne applications. F M 3 T R will allow communications systems to be more affordable and be able to potentially bridge the interoperability gap between nations. The primary goal in this information exchange is improvement in the state of the art

3 Model based Design and SDR

Complexity of software dened radio (SDR) applications and the difculty of testing, debugging and validating of such systems, makes them particularly suitable for development with

Fig. 3. F M 3 T R modulator implemented in Simulink. (courtesy of The MathWorks)

Fig. 5. Test bench to test the compatibility between developed modulator and demodulator. (courtesy of The MathWorks) 2. Differential decoder. Demodulator was also simulated and tested in Simulink. To test the compatibility between the modulator and the demodulator, a test bench was developed using Simulink as shown in Fig.5. To make sure that the modulator and the demodulator are compatible with each other, the output of modulator was fed to the input of demodulator. By comparing the input signal to modulator and the signal out of the demodulator, the compatibility of modulator-demodulator was veried. The modulator was fed by a Bernoulli-distributed random binary number generator. Simulation showed that the designed modulatordemodulator were matched and the error between modulator input and demodulator output were zero.

Fig. 4. F M 3 T R demodulator implemented in Simulink. (courtesy of The MathWorks) of multifunctional radios, primarily focused on hardware and software architectures, digital signal processing, man-machine interfaces, RF and digital technology and interconnecting networks.

6 Code generation
SMT6050 was used for code generation. SMT6050 generates not only all of the necessary C source code but also some other valuable les such as: The linker command le that instructs the linker how to place objects codes into DSP memory. The generated linker command le is compatible with targeted hardware and no user intervention is needed for this purpose. Make le to compile and link the generated source codes. The compiler and linker switches are set according to target hardware and target DSP without any need for user knowledge of underlying hardware architecture and how it would affect the developed software. Batch le to set-up required operating system environmental variable and commands for running make utility. After code generation, SMT6050 runs the generated batch le to build the output le. It also creates required libraries for Simulink block-sets, which are used in the designed model. This created library is compatible with the target hardware and DSP. As one needs to test the generated code and as already

5 Building modulator-demodulator in Simulink

Modulator and demodulator were built and simulated in Simulink. The modulator is show in Fig.3. It was built using standard Simulink blocks. Modulator consists of three sub-systems: 1. Continues-phase encoder. 2. Waveform generators. 3. Memory-less modulator. After building the modulator model in Simulink, it was simulated and its functionality was tested. The demodulator was also built using standard Simulink blocks. It is shown in Fig.4. The demodulator consists of two sub-systems: 1. Optimum state sequence receiver.

Fig. 7. Test bench model prepared for hardware in the loop simulation and testing.

Fig. 6. Hardware in the loop demo that shipped with SMT6050. there was a test bench generated in Simulink, it was logical to use that for testing the application code.

7 Hardware in the loop simulation and testing

In normal simulation all parts of simulation takes place on the Fig. 8. Modulator-demodulator block prepared for hardware in host computer (in case of Simulink, all of the simulation takes the loop code generation. place on the PC). In Hardware In the Loop (HIL) simulation, part of the simulation takes place on the host and another part is executed on the DSP board [6]. SMT6050 includes a demo signals would be in phase with each other. The generated and that shows how to build and test a HIL system. In that demo, processed signals were shown on a scope. data is captured from a microphone on PC and is sent to DSP for processing. DSP processor adds reverb to the sound and 7.2 Modulator-demodulator returns it back to the PC. The processed sound is then played on PC speaker (Fig.6). This is the part that went through code generation and runs To test the generated code for F M 3 T R modulator-demodulator, the DSP. It consists of modulator-demodulator as shown in on the Simulink model shown in Fig.5 was broken into two parts: Fig.8. The Sundance blocks From PC and To PC were added to the model so the input of modulator came from PC and output of demodulator was sent to the PC. 7.1 Test bench Test bench would be run on PC and would feed test vectors to the DSP and then receive the result back from the DSP. By comparing the received signal to the feed signal, the validity of generated code could be veried as shown in Fig.7. The test bench model shown in Fig.5 was used to construct hardware in the loop test bench. Modulator-demodulator were removed from the model and SMT310 block was placed instead of them. This block would then send data to DSP and then receive data from DSP after processing. Some blocks were added to convert data types that would be transmitted to/received from the DSP. Since modulator-demodulator generate 27 time step delays, signal generated by Bernouli binary generator was delayed by 27 so both the original and processed 7.3 Synchronization Since the two different part of this simulation is running on different processor (test bench model on PC and modulatordemodulator on DSP), they should be synchronized with each other [11]. To solve this complex problem, SMT6050 uses data ow synchronization technique. In this technique, the programs on PC and DSP are running freely with their own clock. Test bench sends enough data to DSP so it can start its processing and wait for results to come back from DSP. Modulatordemodulator, which are running on DSP waits until enough data become ready in its input. At this point, DSP starts to process data as fast as it can and sends the resulting data back

awarded the prestiges SMART:SCOTLAND for his innovative technique for image generation from phased array systems with ultrasound, sonar and radar. Mansour is a member of IEE. Zhila (Jila) Nazari is working toward her PhD in the University Of Edinburgh. Her interest is digital signal processing and medical image processing. She received a BSc. and an MSc in Computer software engineering. She developed several commercially available software applications.

Fig. 9. Testing of modulator-demodulator using HIL technique.

Nory Nakhaee received a B.Tech degree in Automotive Engineering and design and a Masters degree in Information to PC. SMT6050 generates all of the required source code for Technology from Loghborough University of TechnolDSP to implement data ow synchronization. It also provide ogy in UK. Nory received a Ph.D. degree from NottingSMT310 block for Simulink so that data communication with ham Trent University in UK for a research titled AutoDSP and synchronization become as simple as possible. matic parallelising complier for MIMD/DF machines. From 1982, for ve years Nory worked at 3L Limited in Scotland as software engineer, test engineer, commercial 7.4 Testing manager and then as a company director. Subsequently To test the validity of the design, code for modulator-demodulator Nory founded Alpha Data Parallel Systems Ltd in Scotis created and downloaded into DSP board. By running test land and served as its Managing Director. In 1999 Nory bench model and comparing the DSP input and output sigmoved to US and founded Sundance DSP Inc and since nals, the validity of modulator and demodulator were veried then he has been working for that company as its C.E.O. as shown in Fig.9. Zoran Kostic obtained a PhD. degree in Electrical Engineering from University of Rochester and Dipl. Ing. De8 Conclusion gree from University of Novi Sad in Yugoslavia. He has worked for Bell Labs, AT&T Research, Thomson ReModel based system design is introduced and the capability of search, The MathWorks and taught at Columbia, UCLA, this methodology is investigated. An FM3TR reference waveSUNY and GMU. His expertise is in the eld of wireform modulator-demodulator was developed using model based less communications research, design and SW and HW methodology and tested. SMT6050 was used to generate code implementation across areas of systems, networks and for hardware in the loop simulation and testing. The generated components. code was tested using hardware in the loop technology. The result of this investigation shows that the application could be easily modeled and then developed without the problems asso- Acknowledgements ciated with traditional design schemes. The authors would like acknowledge the help that they received during this work from MathWorks Inc. specially Amon Gai. Authors also would like to acknowledged helps from their col9 Biography leagues in Sundance, specially Flemming Christensen, Stephen Mansour Ahmadian is the senior software engineer in Sun- Malchi and Sebastien Maury. dance. His current focus and interest is on new development to use model based design for complete embedded References (hardware/software) design. He worked as software developer and recently as software project manager in dif[1] Mansour Ahmadian, Nory Nakhaee, and Andrew Nesferent organizations. terov. Rapid Application Development (RAD) and code optimization technique. Global Signal Processing ConHe received a B.Sc. in Electronic Engineering in 1989, ference (GSPx), 2004. and an M.Sc. in Biomedical Engineering in 1992. He was awarded PhD by the University of Edinburgh for his work in the eld of medical signal processing in 2002. Mansour published more than 20 papers in national and international conferences and journals. In 2001, Mansour won the Business plan competition in the university of Edinburgh. Two years later, he was [2] M. Baleani, A. Ferrari, L. Mangeruca, A.L. SangiovanniVincentelli, U. Freund, E. Schlenker, and H.J.Wolff. Correct-by-construction transformations across design environments for model-based embedded software development. Proceedings of Design, Automation and Test in Europe, 2:1044 1049, 2005.

[3] http://www.mathworks.com. [4] http://www.sundance.com. [5] J. Langenwalter. Embedded automotive system development process - steer-by-wire system. Proceedings of Design, Automation and Test in Europe, 1:538 539, 2005. [6] Jim Ledin and Mike Dickens. Cracking the code automatically. Electronic design europe, pages 19 23, June 2005. [7] J. Loyall, Jianming Ye R. Shapiro, S. Neema, N. Mahadevan, S. Abdelwahed, M. Koets, and D. Varner. A case study in applying qos adaptation and model-based design to the design-time optimization of signal analyzer applications. IEEE Military Communications Conference (MILCOM), November 2004. [8] Arun Mulpur. Faster and better embedded signal processing systems:. Global Signal Processing Conference, 2004. [9] S. Neema, T. Bapty, J. Scott, and B. Eames. Signal processing platform: a tool chain for designing high performance signal processing applications. Proceedings of SoutheastCon, pages 302307, April 2005. [10] T. Runolfsson. Optimal design of uncertain complex dynamical systems. 43rd IEEE Conference on Decision and Control, 2:2231 2236, December 2004. [11] N. Scaife and P. Caspi. Integrating model-based design and preemptive scheduling in mixed time- and eventtriggered systems. Proceedings of 16th Euromicro Conference, June 2004. [12] Richard N. Smith. Description of the f m3 tr proposed reference waveform. Technical report, Software Dened Radio Forum, August 2001. [13] The MathWorks, Inc. Real-Time Workshop users manual , 2002. [14] The MathWorks, Inc. Simulink users manual (version 6), 2005.